I'm trying to set up a pivot_root hack to do some things, switch root,
and then unmount the original root. However, the unmount fails
because ksoftirqd/0, events/0, kblockd/0 and aio/0 have /dev/console
opened. Why are they doing this? Can it be prevented? This happens
when using kernel 2.6.3 (2.6.4 is reportedly broken on Alpha). It
works with a 2.4 kernel using the same script. Does anyone have a
hint?
--
M?ns Rullg?rd
[email protected]
[email protected] (M?ns Rullg?rd) wrote:
>
> I'm trying to set up a pivot_root hack to do some things, switch root,
> and then unmount the original root. However, the unmount fails
> because ksoftirqd/0, events/0, kblockd/0 and aio/0 have /dev/console
> opened. Why are they doing this? Can it be prevented? This happens
> when using kernel 2.6.3 (2.6.4 is reportedly broken on Alpha). It
> works with a 2.4 kernel using the same script. Does anyone have a
> hint?
That's a bug.
keventd and friends are currently holding /dev/console open three times.
It's all inherited from init.
Steal the relevant parts of daemonize() to fix that up.
---
25-akpm/kernel/kthread.c | 18 ++++++++++++++++++
1 files changed, 18 insertions(+)
diff -puN kernel/kthread.c~kthread-keeps-files-open kernel/kthread.c
--- 25/kernel/kthread.c~kthread-keeps-files-open 2004-03-14 14:49:21.679226616 -0800
+++ 25-akpm/kernel/kthread.c 2004-03-14 14:57:18.425750128 -0800
@@ -10,6 +10,7 @@
#include <linux/completion.h>
#include <linux/err.h>
#include <linux/unistd.h>
+#include <linux/file.h>
#include <asm/semaphore.h>
struct kthread_create_info
@@ -41,6 +42,21 @@ int kthread_should_stop(void)
return (kthread_stop_info.k == current);
}
+
+static void kthread_exit_files(void)
+{
+ struct fs_struct *fs;
+ struct task_struct *tsk = current;
+
+ exit_fs(tsk); /* current->fs->count--; */
+ fs = init_task.fs;
+ tsk->fs = fs;
+ atomic_inc(&fs->count);
+ exit_files(tsk);
+ current->files = init_task.files;
+ atomic_inc(&tsk->files->count);
+}
+
static int kthread(void *_create)
{
struct kthread_create_info *create = _create;
@@ -50,6 +66,8 @@ static int kthread(void *_create)
int ret = -EINTR;
cpumask_t mask = CPU_MASK_ALL;
+ kthread_exit_files();
+
/* Copy data: it's on keventd's stack */
threadfn = create->threadfn;
data = create->data;
_
Andrew Morton <[email protected]> writes:
> [email protected] (M?ns Rullg?rd) wrote:
>>
>> I'm trying to set up a pivot_root hack to do some things, switch root,
>> and then unmount the original root. However, the unmount fails
>> because ksoftirqd/0, events/0, kblockd/0 and aio/0 have /dev/console
>> opened. Why are they doing this? Can it be prevented? This happens
>> when using kernel 2.6.3 (2.6.4 is reportedly broken on Alpha). It
>> works with a 2.4 kernel using the same script. Does anyone have a
>> hint?
>
> That's a bug.
>
> keventd and friends are currently holding /dev/console open three times.
> It's all inherited from init.
>
> Steal the relevant parts of daemonize() to fix that up.
That did the trick. Thanks for the quick response.
--
M?ns Rullg?rd
[email protected]